Stephen Wolfram

How should we tell AIs what we want? For the past few decades, the Wolfram Language has been developed as a computational communication language to provide a bridge between human goals and computational capabilities. The Wolfram Language is becoming a full symbolic discourse language that can capture the kinds of human intentions we find in laws and contracts. If we want to tell AIs to “be nice to humans,” how should we actually do that? This chapter first discusses how the development of symbolic discourse language can make it possible to express such ideas. It then considers how we might approach building an AI constitution that defines how we want AIs to act and what ethics they should follow. This chapter also discusses some of the fundamental issues around having a complete computational legal or ethical system.

The most elementary way to think about Mathctrtati ca is as an enhance calculator — a calculator that does not only numerical computation but also algebraic computation and graphics. Matltcmatica can function much like a standard calt".1a- tor. you type in a question, you get back an answer. But Mat/tctttadca ga's turthcr I ue an ordinary calculator. You can type in questions that require answers that arc longer than a calculator can handle. For example, Matltcmatictt can giv; you thc numerical value of tr to a hundred decimal places, or the exact result for a numerical calculation as complicated as the result of 3)tI0, (m Fig. 1)